How to make the interpretation of the environment into the input signal of the ANN and the interpretation of its output signal into the methods of some action in this environment?

Question

Neural networks

Good afternoon!
Recently I tried to make a neural network that would control a certain agent in a certain environment. Methods radar(objectCategory); are available to get information from the environment; - finds the category object in the environment, distance(object); - returns the distance to the object, direction(object); - returns the direction to the object.
I'm interested in the question of feeding the return values ​​of these methods to the input of the neural network. How can you build an interpreter?
And a more difficult question, is it even possible to interpret the output signal in such a way that a method like move(object.position); would be executed. turn(object.direction()); and the like? That is, you need to interpret a set of values ​​​​from 0 to 1 into the name of the function and its parameters, moreover, there is some limited set of these names and parameters. If this is possible, then I would also like to hear how this can be implemented.
PS
Yes, I know that neural networks are poorly designed for such things, but this does not negate my curiosity and desire to make some kind of AI on neural networks.

Answer 1

The NS does not operate with commands, the NS operates with incoming signals from the sense organs and signals to the executive organs. You can perform pre-processing and post-processing of incoming and outgoing signals to improve the performance of the NN.
In general, the NN cannot use "I want to go there", "I'll turn around until", etc.
The NS can issue elementary commands (a finite number) such as "go 1 step forward", "turn 5 degrees to the left", i.e. using feedback from the environment NN + feedback from the "goal-setting function" NN can determine the necessary actions to achieve the goal.